Diagram-MMU - Code Editing: leaderboard
The code-editing track of Diagram-MMU (arXiv 2608.12262): given a diagram’s TikZ source and an instruction, the model must apply the edit and leave the rest of the figure intact; scored on the re-rendered result, reported as the track’s "All" column.
Metric: All Score (%). Source: vi-ocean.github.io. Saturation forecast: Around 2028. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B | 52.25 | #238 |
| 2 | Kimi K2.5 | 52.13 | #144 |
| 3 | Gemini 3.1 Pro (Preview) | 51.22 | #58 |
| 4 | Claude Opus 4.6 | 51.18 | #63 |
| 5 | Qwen 3.5 397B A17B | 50.82 | #146 |
| 6 | GPT-5.2 | 46.15 | #110 |
| 7 | Seed 2.0 Pro | 43.73 | #118 |
| 8 | InternVL3-38B | 38.85 | #405 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=diagram-mmu-code-editing · How It Works · Data refreshed daily, snapshot 2026-10-20.